Research On Spatial Semantic Understanding Of Human Environment For Robot Intelligent Service

At present,robot technology for performing tedious and monotonous service tasks,such as sweeping robots,is relatively mature,but the level of intelligence for performing service tasks needs to be improved.When the robot performs intelligent service tasks,the robot needs to understand the environmental action of the service object and provide corresponding services.For example,when there is a book next to the service object,it can be inferred that the service object is reading a book and provides a comfortable service of pouring water.When the surrounding environment of the service object is complex,the robot also needs to provide emergency services,informing the service object of stopping the activity or removing the obstacle,for the service object based on the information of the environmental space,such as the presence of obstacles or dangerous objects.Therefore,when the robot provides intelligent service tasks,the simple action recognition cannot meet the requirements,and the environment space semantic understanding of the environment where the service object is located must be performed.Therefore,studying the semantic understanding of the human environment space in the indoor environment is the key to accomplishing intelligent service tasks.This topic recognizes the action of the service object in the indoor environment,and obtains environmental information,constructs a human environment semantic logic space containing the service object action and environmental information,and realizes the autonomous reasoning service task of the robot.Firstly,based on the three-dimensional convolutional neural network,the simple recognition of human action is achieved,and through the feature fusion and the transformation of the three-dimensional convolution structure,the effect of action recognition is optimized.Then for the environment where the service object is located,the visual relationship is obtained through the visual relationship detection model,and the action recognition results are synthesized to construct a semantic logical space of the human environment based on the knowledge graph,and the service robot models the human environment space in the indoor environment.The task is reasoned by the service robot based on the semantic logic space of the human environment.(1)Realizing simple and fast action recognition based on three-dimensional convolutional neural network,and improving the accuracy of action recognition through network structure optimization and feature fusion methods.Considering the limited computing power of service robots,C3D network with a simple structure and good real-time performance was selected as a feature extraction network for action recognition.The spatio-temporal features of human action extracted through the C3D network were used for action recognition to achieve simple and rapid recognition of human action.Then the segmented image containing only the object to be recognized was extracted based on Faster R-CNN detection results of the human body in the original image data and performed feature extraction.The results of action recognition by using the two extracted spatiotemporal features were optimized.(2)Optimizing 3D convolutional neural network to improve real-time performance.It was proposed to convert the three-dimensional convolution into a pseudo three-dimensional convolution.The pseudo three-dimensional convolution is composed of a two-dimensional convolution and a one-dimensional convolution,which reduces the calculation amount of the three-dimensional convolution operation and the number of convolution kernel parameters.Using pseudo three-dimensional convolution operation replaced part of the three-dimensional convolution operation in the network model achieved the purpose of reducing network model parameters and calculation costs,and further optimizes the effect of action recognition.Real-time performance was greatly improved.(3)Constructing the semantic logic space of human environment and deeply understanding the environmental action of service objects.Aiming at the intelligent service requirements of service robots,we deeply understanded the environmental action of service objects.The visual relationship of the environment where the service object is located was obtained based on the visual relationship detection model combined the results of action recognition to construct the human environment semantic logical space which is based on the knowledge graph.Deep action of the service object was obtained via using the knowledge graph relationship search and rule reasoning.Then a deep understanding of the action of the service object and the environment was achieved and it was achieved that service robots reason intelligent service tasks independently.The method proposed in this paper fully considers the impact of complex environment on the action of the service object and models the human environment space in the indoor environment semantically.It is of great significance for robots to provide intelligent service tasks better.The experimental results show that the action recognition algorithm used in this paper has good recognition accuracy and real-time performance and the semantic logic space of human environment proposed in this paper can be modeled through behavior recognition and environmental information acquisition and can be used for autonomous service inference,which verifies the feasibility and effectiveness of the method in this paper.